2023-01-19: APP 2.0.0-beta13 has been released !
!!! Big performance increase due to optimizations in integration !!!
and upgraded development platform to GraalVM 22.3 based on openJDK19
We are very close now to releasing APP 2.0.0 stable with a complete printable manual...
Quality score caculation
How is the quality score caculated (reading the other posts I only get half answers about it being relative, but no simple (or complicted :)) formula that Mabula has promised earlier….
I am really curious!
I too have been curious about the calculation of quality for somewhile and believe I have read most if not all that Mabula has had to say on this matter. So for my benefit as much as yours I offer these thoughts on your question that you and others and may take, leave or amplify as you think best.
I suggest we can probably mostly agree that major factors that help contribute to a good sub are:
- Good seeing:
- Good tracking:
- Good focus
Thus if we were to compare several subs of the same target, taken with the same filter and covering the exact same area of sky, we would likely decide that the better subs were the ones having:
- The highest star count - these will generally be the ones taken when the sky was most transparent
- The least elliptical (i.e, roundest) stars - these will likely be the ones taken when the imaging system was tracking at its best
- The smallest stars - these would likely be the ones taken when the imaging system was closest to best focus.
In his write up of the changes he has made for release v1.076 Mabula states:
'the quality calculation for frames after Star Analysis is now much more robust, the formula is now: numberOfStars * median star size * median star roundness. The star size and roundness are based on the median star profile of all stars analysed. Previously, the calculation was much more subject to outliers, because the calculation was a summation of star size and roundness per individual analysed star.'
Some of the reasons why the quality measure is relative and thus specific only to a particular set of frames of a given target and filter are:
- If you image different areas of sky with the same imaging system you will almost certainly capture different numbers of stars in the frame;
- If you image closer to the horizon the photons must travel through a greater thickness of atmosphere and thus subject to more atmospheric turbulance and attenuation, thus impacting star roundness and star count;
- Imaging with different filters will affect photon counts and thus star counts.
: This is to say that subs taken of Target A but obtaining only a lowish quality score are not necessarily significantly worse in overall quality terms when compared to subs taken of Target B which obtained a higher quality score. But, as we would not be integrating Target A subs with Target B subs, these quality score differences are not of significance. APP is only interested in the relative quality rankings of the Target A subs while it is integrating Target A. Though if we are using quality to weight our integration then it is important that the quality scores strongly correlate to the actual quality of the subs relative to each other. I have occasionally wondered if other easy to obtain measures might also be included within the quality algorithm but have not come up with anything much apart from perhaps contrast range but this probably correlates quite closely with 'star count' so would not offer much added value. Also is there a way to quantify light pollution gradients and reduce the quality score for a frame if these are significant?
I hope this idle write-up is somewhat useful in promoting understanding and that I am not too far off-beam with my explanation.
You're spot on there Mike. Indeed these scores are completely relative, they are not to be seen as an overall score for sky quality, they are calculated on the basis of the dataset you provide. Thanks Mike.
I'm testing APP coming from DSS and Siril and I'm really liking it! One thing I have just been surprised about is the quality score.
I'm trying to integrate several sessions on the Lagoon Nebula, and one of them only have dark frames but no flats nor bias, as the rest of the sessions.
For that particular session, APP 1.082 reports a Q score of ~3000 and on the rest the score goes around 500-700. Is this a sign that my calibration frames are affecting negatively on the entire process?